nRF51822 32 kHz crystal problem: Connection is lost when supply voltage (VDD and AVDD) is increased by an external DCDC converter

[puz_md]

Hello everybody,

We are using a nRF51288 with an external DCDC converter. VDD and AVDD is bridged, DCC is not connected. The software is working in Peripheral mode, S110 SoftDevice 7.1.0 is used. We use a connection interval of ~30 ms and a timeout of 4 seconds.

When a certain GATT value is written to, we use the BLE_GATTS_EVT_WRITE event handler to increase the operating voltage from 1,8V to 2,8V (which takes ~300-400 µs) and do some I/O , decrease operating voltage again and return. Just in the moment when the voltage is increased, the peripheral stops replying to the connection events from the Central (this was observed with a sniffer). The Peripheral does not even send the acknowledgement for the "Write Request". After a while, we get a connection timeout. (No error handler or system reset is invoked!) In some cases, just a moment before the timeout would happen, there may be a re-synchronisation of the connection.

Does anyone have an idea what could prevent the SoftDevice from properly keeping the connection alive? Can this be caused by the rising operating voltage? Or is there a time limit for returning from the BLE_GATTS_EVT_WRITE event handler?

This is a major problem that could ruin our whole system design... I hope somebody can give some advice (or at least explain the reason for the problem).

Thanks a lot!

Update (Feb 27, 2015):

I tracked down the problem to the 32 kHz crystal: The voltage switching seems to disturb the time base used for Bluetooth. When I set the crystal precision to 500 ppm in the SoftDevice initialization routine, the connection is re-synchronized after approximately 0,8 seconds. Our I/O operation (which needed the higher operating voltage) finished successfully.

• Can anyone of the Nordic employees confirm this behavior?
• Can this be considered a hardware problem? I could not find any constraints on the Bluetooth operation related to the supply voltage.
• How is the Bluetooth / connection stability if the internal DCDC converter is used and the external voltage rises or falls abruptly?
• Is there any way to prevent the connection loss?
• Is there any way to dynamically set the 32 kHz precision setting used by the SoftDevice? E.g. increase the tolerance (to a higher value than 500 ppm) when voltage switching takes place, and decrease it after the voltage has been switched back and the crystal has stabilized again.

As our project is in its final stage we need quick response.

Best regards

[Asbjørn]

• Changing the voltage abruptly could lead to scenarios where crystal or other parts of the design gets affected in a non-deterministic way. I would suggest to scope on the VDD pin and on the crystal pin during the voltage change. If you are able to identify abnormal activity when the voltage change occurs I would recommend to add some capacitance on the VDD pins and DCDC to filter out spikes/spurs. The voltage change itself could also change to run time of the oscillator. So make the voltage change slower would be a benefit.

• I would consider it a HW problem as I'm sure that it's possible to find a solution on how to smooth out the voltage changes.

• It goes through different regulators and is decoupled so it doesn't affect the oscillator circuit.

• Switching 32 kHz settings can only be done by turning SoftDevice on/off.

[puz_md]

Thanks for the feedback. In the meantime, we increased the external voltage in steps of 0.1V, each step is nrf_delay_ms(1) long. This was, the connection is not lost during our lab tests (we don't know if connection quality will be effected in field operation, though).

We already used sufficient capacitance on the VDD pins, there were no spikes visible. We also couldn't detect suspicious behavior on the 32 kHz crystal (there was a slight offset shift during voltage switching, but the frequency appears to be stable).

Is there any official information available under which VDD conditions the radio circuit will work reliable? I could not find any information about acceptable voltage drifts in the nRF51822 specification.